Our paper is currently under review
Our model adapts to both pose and garment features, providing a high-quality virtual try-on experience.
Below is the pipeline of our model, detailing each step from input to output:
Here are the results after 10 hours of training (20 epochs) using a single A100 GPU:
To prepare the "In-shop Clothes Retrieval Benchmark" dataset, follow these steps:
- Download the dataset from DeepFashion: In-shop Clothes Retrieval Benchmark.
- This dataset includes:
- 7,982 clothing items.
- 52,712 in-shop clothes images.
- Approximately 200,000 cross-pose/scale pairs.
- Each image is annotated with bounding box, clothing type, and pose type.
- Extract the downloaded files into the
Fashionfolder within your project directory to maintain the required structure.
Guideline & CSV file for Fashion Tryon Dataset will be provided later
- Checkpoint update
- Training scripts with detailed usage instructions
- Scripts for ablation studies
- Demo
-
Model pipeline
To set up and run our model, follow these steps:
- Clone the repository:
git clone https://github.com/seohyun8825/ODPG_1.git - Install required packages:
pip install -r requirements.txt
This project is built on top of the CFLD official code. The original codebase has been modified to include additional conditioning on garment features, enabling the model to handle more complex virtual try-on scenarios where both pose and clothing attributes are considered.